OpenAI researcher discusses the technical challenge of aligning superintelligent AI systems that will vastly exceed human capabilities. Current alignment techniques like RLHF won't scale to superhuman systems, posing risks during rapid intelligence explosions, though the author believes solutions exist with sufficient effort and empirical progress.